SANTA CLARA, CA -- Sensory Inc., a Silicon Valley based company focused on machine learning techniques to improve the user experience and security of consumer electronics through state-of-the-art embedded voice and vision technologies, announces the release of TrulyNatural, a breakthrough client-based large vocabulary speech recognition platform that offers substantive advantages over cloud-based solutions from the industry giants, while offering industry-leading accuracy, and a more consistent and secure user experience. Additionally, Sensory announced today that it has signed Jibo, Inc., creators of the world's first social robot for the home, as its first licensee of TrulyNatural.
"Over the years, Sensory has pioneered many speech technology trends, most recently with our always-on, always-listening speech solution TrulyHandsfree, a technology that many in the industry thought was impossible to achieve because of the difficult requirements of both high accuracy and low power consumption. We are confident that we have another ground-breaker on our hands with TrulyNatural," said Todd Mozer, CEO of Sensory. "Working with Jibo, Inc. as our premiere customer for TrulyNatural is particularly exciting because they have some of the world's leading speech technologists on their team. We know TrulyNatural will be in the hands of experts who can fully utilize the many advantages of our breakthrough approach!"
Sensory uses a unique form of a neural network with deep learning to achieve acoustic models an order of magnitude smaller than the present state-of-the-art. These neural networks employ the most recent breakthroughs in speech feature extraction to produce superior accuracy in real world noise. Combining this with a very efficient Finite State Transducer (FST) enables accurate processing of multiple large search domains as required in today's marketplace.
"We wanted a natural language interface that would just work, and we didn't want to be tied to a cloud connection. The Jibo team was familiar with Sensory's excellent reputation from their TrulyHandsfree approach, and when we learned about TrulyNatural, it became an easy choice to work with Sensory," said Cynthia Breazeal, Founder and Chief Scientist of Jibo, Inc. "Sensory's technologies have a reputation of strong performance and accuracy. We find it particularly comforting that Jibo, the world's first social robot for the home, can process all speech data received by its users internally, opposed to having to be reliant on a strong and stable internet connection for cloud processing of speech."
Custom designed and developed at Sensory with over 50 man-years effort by a team of speech technologists, TrulyNatural was designed to be:
•Natural – capable of allowing users to speak to their devices naturally as they would to another person
•Consistent – the solution works exactly as it should with or without an internet connection
•Scalable – ideal for smaller footprint device requiring vocabularies of a hundreds of words in under 1 MB of memory to smartphones, cars, and robots that require more natural language interfaces that still fit on device
•Fast – with search taking place on the client device instead of the cloud, requests are processed without delay. Because Sensory has focused on reduced MIP implementations, the result is a faster response time with any given power level
•Accurate – delivers an industry's best <8% word error rates (WER) for 1 million entries using native speakers. Better than many published benchmarks for cloud based solutions
•Flexible – TrulyNatural can be used in combination or without cloud based solutions, and can take industry standard components for language models to make development faster and easier. Grammars and language can be used separately or together
TrulyNatural is initially released for US English, Mandarin Chinese, Korean, Japanese, UK English, French, Spanish and German, with Italian, Portuguese, and Russian to follow this year. SDK's are available for Android, iOS, Windows, Linux and other leading platforms. Jibo robots are planned to be demonstrated in 2015 and released in 2016 with a TrulyNatural natural language interface.